Research On Detection And Defense Methods Against Load Integrity Attack On Smart Grid

The smart grid as a national infrastructure provides reliable, inexpensive and sustained power services. Accurate state estimation is of paramount importance to maintain normal operations of smart power grids. However, a new type of attacks(called the load integrity attack, LIA) has been introduced, which can bypass the bad data detection (BDD) systems, make the control center do a series of wrong decisions, and result in the imbalance of power grid load distribution. In smart grid environment, this paper analyzes the attack mechanism and proposes new-type attack. So it will be of important significance to study the corresponding detection methods countermeasures for smart gird.The paper concentrates on the LIA problem in smart grid, including:1. This paper introduces the basic principle of LIA and studies the general problem of attack mode without the knowledge of power network topology; We maximize the approximation of Jacobian matrix using principal component analysis(PCA) method to construct the attack vector; calculate the relationship between the measurement and the independent component of the state variables using independent component analysis(ICA). In this way, attackers can launch two new-type attacks.2. Considering the attack vector, this paper constructs illegal measurements, and reduce dimensionality of the generated data by representing high-dimensional data in a low dimensional space using Autoencoder. The simulation experimental results about reconstruction error in pre-training and fine-tuning process, which shows that low-dimensional data information received by reducing dimensionality of data can basically fully represent high-dimensional data.3. From the respective of algorithm optimization, this paper proposes the improved extreme learning machine(ELM) and applies it into attack detection. The differential evolution algorithm(DE) incorporates the thought of artificial bee colony(ABC) in the optimization algorithm for premature and slow convergence in DE, which will help DE escape the local optimum. Moreover, the detection precision and the generalization capability of the proposed method is greatly improved.4.We demonstrate the effectiveness of our detection by a lot of experiments in IEEE 14-bus and 118-bus systems, respectively. In the end, the defense mechanism has been proposed based on the protection sets method.